A significant number of artificial intelligence startup concepts remain essentially superficial "wrappers" constructed atop existing foundational models. However, as the core AI model developers integrate more capabilities directly, investors are growing cautious of startups that could be rendered obsolete so easily. This trend was evident in the review of over 4,000 applications for a joint AI accelerator program for Indian startups, operated by Google and venture firm Accel, where "wrapper" ideas were predominant.

Announced in November, the AI-focused Atoms program by Google and Accel is designed to support early-stage startups developing AI products with a connection to India. Selected startups for the latest cohort will obtain up to $2 million in funding from Accel and Google’s AI Futures Fund, in addition to up to $350,000 in cloud and AI compute credits from Google.

According to the firms, approximately 70% of the rejected applications were classified as "wrappers." These were startups that merely added AI features, like chatbots, to existing software without fundamentally "reimagining new workflows using AI," as Swaroop noted. Many of the other denied applications fell into saturated categories such as marketing automation and AI recruitment tools, areas where investors perceived little innovation. Startups in these sectors frequently find it challenging to stand out, he added.

This outcome is not entirely unexpected. The current program attracted nearly four times the number of applications compared to previous Accel Atoms cohorts, including many first-time founders. India's expanding AI ecosystem continues to be heavily oriented toward enterprise applications, a focus clearly reflected in the submissions, Swaroop stated. About 62% of the proposals centered on productivity tools, with another 13% on software development and coding. This means roughly three-quarters of all applications were for enterprise software ideas rather than consumer products. (Swaroop expressed a desire to have seen more concepts targeting healthcare and education.)

Jonathan Silber, co-founder and director of Google’s AI Futures Fund, mentioned that the five chosen startups align closely with domains where Google anticipates deeper, real-world adoption of AI. He clarified that the program does not mandate exclusive use of Google’s models, noting that many companies integrate multiple models based on specific workflow needs. The objective, Silber explained, is to collect feedback from startups on the performance of Google’s models in practical applications. Insights gained from these startups can then be channeled back to Google DeepMind teams to aid in refining future models, establishing what Silber termed a "flywheel" effect between startup experimentation and AI development.

The selected startups for this year are:

* K-Dense, which is developing an AI "co-scientist" to speed up research in fields like life sciences and chemistry; * Dodge.ai, which creates autonomous agents for enterprise ERP systems; * Persistence Labs, which specializes in voice AI for call center operations; * Zingroll, which is building a platform for AI-generated films and shows; * Level Plane, which applies AI to industrial automation within automotive and aerospace manufacturing.